Meta AI

Overview

Meta’s AI division (formerly FAIR — Facebook AI Research) has become the champion of open-weight AI models. The LLaMA model family is the most widely used open foundation model, powering thousands of applications and fine-tunes worldwide.

Key Models

• LLaMA 4 — Latest open-weight models (2025)
• LLaMA 3.1 (405B/70B/8B) — Widely adopted open models
• Code LLaMA — Code-specialised models
• Segment Anything (SAM) — Vision segmentation
• ImageBind — Multimodal embeddings

Products & Services

• Meta AI Assistant — Consumer AI across WhatsApp, Instagram, Facebook
• LLaMA models — Open-weight downloads for anyone to run/fine-tune
• PyTorch — The dominant deep learning framework (Meta-maintained)

Key People

• Yann LeCun — Chief AI Scientist (Turing Award winner)
• Mark Zuckerberg — CEO (personally champions open AI strategy)
• Ahmad Al-Dahle — VP of Generative AI

Notes

• Open-weight strategy: gives models away to build ecosystem, undercut competitors
• LLaMA download numbers in the hundreds of millions
• PyTorch is the de facto standard for AI research and production
• Massive compute investment ($30B+ annually on AI infrastructure)
• Philosophy: AI should be open and decentralised (LeCun vocal critic of “AI doomerism”)